Anyone here with 2TB or 4TB drives at Hetzner?

Amitz

I wonder if anyone here is having a server @ Hetzner with 2TB or 4TB drives in Software Raid-1 and would be willing to share the IO that those drives are capable to deliver?

The infamous "dd" test and (if possible) "ioping" results would be just great. I already have servers with their 3TB drives and would love to see whether the other models are a tad faster.

Thanks a lot in advance & kind regards
Amitz

P.S.: For comparison, here are the results of the 3TB drives

dd bs=1M count=512 if=/dev/zero of=test conv=fdatasync && rm -f test
512+0 records in
512+0 records out
536870912 bytes (537 MB) copied, 5.56555 s, 96.5 MB/s

dd bs=1M count=512 if=/dev/zero of=test conv=fdatasync && rm -f test
512+0 records in
512+0 records out
536870912 bytes (537 MB) copied, 5.67656 s, 94.6 MB/s

dd bs=1M count=512 if=/dev/zero of=test conv=fdatasync && rm -f test
512+0 records in
512+0 records out
536870912 bytes (537 MB) copied, 5.59846 s, 95.9 MB/s

ioping -c 50 /  
--- / (ext4 /dev/md2) ioping statistics ---
50 requests completed in 49.1 s, 693 iops, 2.71 MiB/s
min/avg/max/mdev = 195 us / 1.44 ms / 19.1 ms / 4.01 ms

ioping -c 50 /
--- / (ext4 /dev/md2) ioping statistics ---
50 requests completed in 49.0 s, 2.46 k iops, 9.62 MiB/s
min/avg/max/mdev = 195 us / 405 us / 7.36 ms / 994 us

ioping -c 50 /
--- / (ext4 /dev/md2) ioping statistics ---
50 requests completed in 49.1 s, 811 iops, 3.17 MiB/s
min/avg/max/mdev = 198 us / 1.23 ms / 15.4 ms / 3.16 ms

rds100

What 3TB drives are these? 7200RPM or 5400/5900RPM? Are your partitions aligned on 4K?

rm_

To get meaningful ioping results, use ioping -R /.

loot

2 x 2 but in LVM

dd bs=1M count=512 if=/dev/zero of=test conv=fdatasync && rm -f test
12+0 records in
512+0 records out
536870912 bytes (537 MB, 512 MiB) copied, 2.08897 s, 257 MB/s

dd bs=1M count=512 if=/dev/zero of=test conv=fdatasync && rm -f test
512+0 records in
512+0 records out
536870912 bytes (537 MB, 512 MiB) copied, 2.06934 s, 259 MB/s

dd bs=1M count=512 if=/dev/zero of=test conv=fdatasync && rm -f test
512+0 records in
512+0 records out
536870912 bytes (537 MB, 512 MiB) copied, 2.09373 s, 256 MB/s

--- / (ext4 /dev/mapper/vg0-root) ioping statistics ---
50 requests completed in 49.1 s, 409 iops, 1.60 MiB/s
min/avg/max/mdev = 224 us / 2.44 ms / 21.4 ms / 5.53 ms

--- / (ext4 /dev/dm-0) ioping statistics ---
50 requests completed in 49.1 s, 741 iops, 2.90 MiB/s
min/avg/max/mdev = 201 us / 1.35 ms / 15.9 ms / 3.13 ms

Both are 2x2 and LVMed

vfuse

4TB drives not in RAID1 tho (jbod)

dd bs=1M count=512 if=/dev/zero of=test conv=fdatasync && rm -f test
512+0 records in
512+0 records out
536870912 bytes (537 MB) copied, 5.28714 s, 102 MB/s

--- / (ext4 /dev/disk/by-uuid/b3b14b31-5aca-4a18-8379-9c80876be454) ioping statistics ---
50 requests completed in 49.2 s, 316 iops, 1.2 MiB/s
min/avg/max/mdev = 66 us / 3.2 ms / 81.6 ms / 12.0 ms

Amitz

Thank you, guys!

ehab

didn't read all but did you try enable write cache

hdparm -W /dev/sdX

replace X with what you have and run tests again

Amitz

@ehab said:
didn't read all but did you try enable write cache

hdparm -W /dev/sdX

replace X with what you have and run tests again

No change. I guess that write cache was already enabled...
But to be precise - It's not that I am unhappy with the disk speed. I just wanted to know whether a model change would also bring a benefit concerning disk IO. Obviously not (or at least not significantly).